Hi,
I am trying to determine the best arrangement for setting up the disk
subsystem on a Win2k/Oracle 8i EE configuration. The db will see a 50/50 mix
of OLTP and batch processing/reporting usage. Emphasis is on making the OLTP
side perform. The system is a dual PIII with 2G of RAM, a dual-channel RAID
controller with 256 Mb of cache RAM and 10 fast, 15000 RPM hot-swap disk
drives. The database size will be in the 10 to 20 Gb range. It is a
requirement that Oracle must run in ARCHIVE mode.
I am considering the following:
- Allocating three of the pairs in RAID 10 configuration
On it, creating three logical volumes:
- A small one striped at 64k for OS+paging
- A main one striped at 64k for the main tablespaces (data+index)
- A support one striped at 256k for rollback and temporary tablespaces
- Allocate the fourth mirrored pair for the archive logs
- Keep the last pair split and have Oracle multiplex the online redo logs on
these drives (for extra protection)
Questions based on your experience (knowing there is no easily transferrable
answer):
- Is Oracle multiplexing sufficient to continue processing in case of
failure of one of the online redo disks?
- Am I really to benefit the most from separating these three I/O types, or
should I combine the online and archive redo I/O on a smaller 2-pair,
4-drive RAID 10? Note that in this setup each RAID channel is to control one
"side" of the arrangement (i.e. three striped pair members, one online redo
disk, one archived redo mirror).
- Would you suggest any alternative placement? maybe striping across 4 pairs
and using the remaining pair for online redo? Assuming 9Gb of 'effective' db
space per drive pair in the main RAID 10 array, does it make sense to
consider using *it* for archive log storage, maybe creating a fourth, large
1 Mb stripe set, dedicated to this?
Thanks for any and all pointers!
Thanassis
Received on Sat Oct 20 2001 - 10:42:02 CDT